Research map / Image generation / Autoregressive and latent generation
MetaDiT: Enabling Fine-grained Constraints in High-degree-of Freedom Metasurface Design
Hao Li, Andrey Bogdanov
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| add pretrained contrastive spectrum encoder (ablation) | MetaDiT · Metasurface inverse design (test set) | AAE: 82.4838 -> 48.2495 | Improved |
| add coarse condition via adaLN (ablation) | MetaDiT · Metasurface inverse design (test set) | AAE: 59.9662 -> 48.2495 | Improved |
| use in-context conditioning instead of cross-attention (ablation) | MetaDiT · Metasurface inverse design (test set) | AAE: 55.8119 -> 48.2495 | Improved |
| use high-resolution spectra as input | MetaDiff · Metasurface inverse design (test set) | AAE: 112.0591 -> 79.1365 | Improved |
| replace vanilla DiT with MetaDiT conditioning | DiT · Metasurface inverse design (test set) | AAE: 100.9437 -> 48.2495 | Improved |
| replace MetaDiff-HR with MetaDiT | MetaDiff-HR · Metasurface inverse design (test set) | AAE: 79.1365 -> 48.2495 | Improved |
| scale MetaDiT to MetaDiT-B (57.78M params) | MetaDiT · Metasurface inverse design (test set) | AAE: modest performance gain of 2.3% | No clear effect |
| scale MetaDiT-B to MetaDiT-L (129.73M params) | MetaDiT-B · Metasurface inverse design (test set) | AAE: performance drop of 15.7% | Hurt |
| scale up model size | MetaDiT · Metasurface inverse design (test set) | validation MSE: consistently lowers the diffusion MSE | Improved |
Closest papers by meaning
- Wyckoff Transformer: Generation of Symmetric Crystals (ICML 2025 · similarity 0.82)
- UniMate: A Unified Model for Mechanical Metamaterial Generation, Property Prediction, and Condition Confirmation (ICML 2025 · similarity 0.81)
- Energy-conserving equivariant GNN for elasticity of lattice architected metamaterials (ICLR 2024 · similarity 0.81)
- Generating Symmetric Materials using Latent Flow Matching (NeurIPS 2026 · similarity 0.81)
- Conformal Crystal Graph Transformer with Robust Encoding of Periodic Invariance (AAAI 2024 · similarity 0.81)
- Co-PiLOT: Constrained Physics-Informed Latent Optimization for Target-Driven Inverse Design (NeurIPS 2026 · similarity 0.80)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
